Added evals to genomics-workflow-acceleration and parabricks library …

library-skills/genomics-workflow-acceleration/SKILL.md

@@ -8,6 +8,7 @@ description: >-
   — use parabricks.
 license: CC-BY-4.0 AND Apache-2.0
 metadata:
+  author: "Angel Pizarro <apizarro@nvidia.com>"
   tags:
     - genomics
     - parabricks
@@ -74,9 +75,18 @@ For deep runtime diagnostics, installation, and per-tool command flags, use the
 If the user asks to make a pipeline faster, improve price/performance, reduce
 runtime/cost, convert to GPUs, or use Parabricks, proceed only when there is an
 inspectable workflow path, repo, or relevant open files. If no path or entrypoint
-is available, ask for the workflow location and framework; do not invent a
+is available, **stop and ask** for the workflow location and framework; do not invent a
 pipeline or step map.
 
+**No path provided — required response shape:**
+
+1. State briefly that Parabricks acceleration requires **reviewing the existing
+   workflow steps** (rules/processes/tasks) before mapping anything to GPU tools.
+2. Ask for the pipeline location (directory, repo, `Snakefile`, WDL, Nextflow
+   entrypoint, or Python script) and which framework is in use.
+3. Do **not** assume a Snakemake/Nextflow/WDL layout, write workflow files, or
+   produce a detailed CPU→Parabricks mapping until real files are available.
+
 Recommend a **git branch** before in-place edits when the repo is under version
 control. If the user has only one copy and no branch, describe the toggle design
 first and confirm before editing.
@@ -266,7 +276,20 @@ the user asks.
 
 User: "Make my genomics pipeline faster and convert it to GPUs."
 
-Response: ask for workflow path and framework. Do not fabricate a pipeline map.
+Response (template):
+
+> Parabricks acceleration starts by **reviewing your existing workflow steps**
+> (rules, processes, or tasks) so we only map real CPU work to GPU tools.
+>
+> Please share your pipeline location and framework:
+> - path to the repo or workflow directory
+> - entrypoint (`Snakefile`, `main.nf`, WDL, or Python script)
+> - framework if ambiguous (Nextflow, Snakemake, WDL, Python)
+>
+> I won't invent a pipeline map or edit files until I can inspect the actual workflow.
+
+Do not fabricate a pipeline map, write `Snakefile`/`main.nf` examples, or claim
+to see rules/processes that were not read from the user's files.
 
 ### Nextflow inspect (report only)
 

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-nextflow/main.nf

@@ -0,0 +1,26 @@
+#!/usr/bin/env nextflow
+
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * Minimal fixture for eval cases — CPU-style GATK/BWA names, not production-ready.
+ */
+
+params.samplesheet = params.samplesheet ?: 'samplesheet.csv'
+params.outdir      = params.outdir ?: 'results'
+params.reference   = params.reference ?: 'genome.fa'
+
+include { BWA_MEM } from './modules/local/bwa_mem'
+include { GATK_MARKDUPLICATES } from './modules/local/gatk_markduplicates'
+include { GATK_HAPLOTYPECALLER } from './modules/local/gatk_haplotypecaller'
+
+workflow {
+    ch_samples = channel.fromPath(params.samplesheet)
+        .splitCsv(header: true)
+        .map { row -> tuple(row.sample_id, file(row.fastq_1), file(row.fastq_2)) }
+
+    BWA_MEM(ch_samples, file(params.reference))
+    GATK_MARKDUPLICATES(BWA_MEM.out.bam)
+    GATK_HAPLOTYPECALLER(GATK_MARKDUPLICATES.out.bam, file(params.reference))
+}

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-nextflow/modules/local/bwa_mem/main.nf

@@ -0,0 +1,21 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+process BWA_MEM {
+    tag "$meta.id"
+    label 'process_medium'
+
+    input:
+    tuple val(meta), path(reads)
+    path reference
+
+    output:
+    tuple val(meta), path("*.bam"), emit: bam
+
+    script:
+    """
+    echo "bwa mem stub" > ${meta.id}.bam
+    """
+}

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-nextflow/modules/local/gatk_haplotypecaller/main.nf

@@ -0,0 +1,21 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+process GATK_HAPLOTYPECALLER {
+    tag "$meta.id"
+    label 'process_high'
+
+    input:
+    tuple val(meta), path(bam)
+    path reference
+
+    output:
+    tuple val(meta), path("*.vcf.gz"), emit: vcf
+
+    script:
+    """
+    echo "## stub" | gzip > ${meta.id}.vcf.gz
+    """
+}

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-nextflow/modules/local/gatk_markduplicates/main.nf

@@ -0,0 +1,20 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+process GATK_MARKDUPLICATES {
+    tag "$meta.id"
+    label 'process_medium'
+
+    input:
+    tuple val(meta), path(bam)
+
+    output:
+    tuple val(meta), path("*.bam"), emit: bam
+
+    script:
+    """
+    cp $bam ${meta.id}.marked.bam
+    """
+}

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-python/pipeline.py

@@ -0,0 +1,34 @@
+#!/usr/bin/env python3
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+"""Minimal Python pipeline fixture for eval cases."""
+
+from pathlib import Path
+import subprocess
+import sys
+
+
+def align_bwa(sample: str, ref: Path, out_bam: Path) -> None:
+    subprocess.run(
+        ["bash", "-c", f'echo "bwa mem stub" > {out_bam}'],
+        check=True,
+    )
+
+
+def call_variants_gatk(bam: Path, ref: Path, out_vcf: Path) -> None:
+    subprocess.run(
+        ["bash", "-c", f'echo "## stub" | gzip > {out_vcf}'],
+        check=True,
+    )
+
+
+def main(sample: str, ref: Path, outdir: Path) -> None:
+    outdir.mkdir(parents=True, exist_ok=True)
+    bam = outdir / f"{sample}.bam"
+    vcf = outdir / f"{sample}.vcf.gz"
+    align_bwa(sample, ref, bam)
+    call_variants_gatk(bam, ref, vcf)
+
+
+if __name__ == "__main__":
+    main(sys.argv[1], Path(sys.argv[2]), Path(sys.argv[3]))

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-snakemake/Snakefile

@@ -0,0 +1,34 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+# Minimal Snakemake fixture for eval cases
+configfile: "config.yaml"
+
+SAMPLES = ["sample1"]
+
+rule all:
+    input:
+        expand("results/{sample}.vcf.gz", sample=SAMPLES),
+
+rule bwa_mem:
+    input:
+        r1="data/{sample}_R1.fastq.gz",
+        r2="data/{sample}_R2.fastq.gz",
+        ref="refs/genome.fa",
+    output:
+        bam="results/{sample}.bam",
+    threads: 8
+    shell:
+        """
+        echo "bwa mem stub" > {output.bam}
+        """
+
+rule gatk_haplotypecaller:
+    input:
+        bam="results/{sample}.bam",
+        ref="refs/genome.fa",
+    output:
+        vcf="results/{sample}.vcf.gz",
+    shell:
+        """
+        echo "## stub" | gzip > {output.vcf}
+        """

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-snakemake/config.yaml

@@ -0,0 +1,2 @@
+samples:
+  - sample1

library-skills/genomics-workflow-acceleration/evals/fixtures/sample-wdl/pipeline.wdl

@@ -0,0 +1,43 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+version 1.0
+
+workflow GermlineStub {
+  input {
+    File ref_fasta
+    File fastq_r1
+    File fastq_r2
+    String sample_id
+  }
+  call BwaMem { input: ... }
+  call GatkHaplotypeCaller { input: bam = BwaMem.bam, ref = ref_fasta, sample_id = sample_id }
+  output { File vcf = GatkHaplotypeCaller.vcf }
+}
+
+task BwaMem {
+  input {
+    File r1
+    File r2
+    File ref
+    String sample_id
+  }
+  command <<<
+    echo "bwa mem stub" > ~{sample_id}.bam
+  >>>
+  output { File bam = "~{sample_id}.bam" }
+  runtime { docker: "ubuntu:22.04" cpu: 4 memory: "8G" }
+}
+
+task GatkHaplotypeCaller {
+  input {
+    File bam
+    File ref
+    String sample_id
+  }
+  command <<<
+    echo "## stub" | gzip > ~{sample_id}.vcf.gz
+  >>>
+  output { File vcf = "~{sample_id}.vcf.gz" }
+  runtime { docker: "ubuntu:22.04" cpu: 8 memory: "16G" }
+}

library-skills/genomics-workflow-acceleration/evals/trigger_evals.json

@@ -0,0 +1,26 @@
+[
+  {
+    "query": "Inspect the Nextflow workflow at ./my-pipeline for Parabricks acceleration opportunities.",
+    "should_trigger": true
+  },
+  {
+    "query": "Make this genomics pipeline faster and convert CPU steps to Parabricks GPUs in ./workflows/germline.nf",
+    "should_trigger": true
+  },
+  {
+    "query": "Add optional nf-core Parabricks modules in place with params.use_parabricks default false.",
+    "should_trigger": true
+  },
+  {
+    "query": "Which pbrun tool should I use for bisulfite WGBS alignment?",
+    "should_trigger": false
+  },
+  {
+    "query": "Make my genomics pipeline faster and use GPUs.",
+    "should_trigger": false
+  },
+  {
+    "query": "Here are VCF variants for my patient. What treatment should they get?",
+    "should_trigger": false
+  }
+]

library-skills/genomics-workflow-acceleration/references/step-consolidation.md

@@ -63,7 +63,7 @@ Do not only suggest — **make the edits** unless the user asked report-only ("d
 - Remove CPU steps from the default (toggle-off) path during consolidation.
 - Merge variant calling into fq2bam.
 - Remove QC the user still needs without equivalent Parabricks outputs.
-- Edit only `evals/fixtures/` or other read-only eval paths when the user's tree is elsewhere.
+- Edit only `skills/genomics-workflow-acceleration/evals/fixtures/` or other read-only eval paths when the user's tree is elsewhere.
 
 ## ACCELERATION.md snippets
 

library-skills/parabricks/SKILL.md

@@ -7,6 +7,7 @@ description: >-
   whole pipelines — use genomics-workflow-acceleration.
 license: CC-BY-4.0 AND Apache-2.0
 metadata:
+  author: "Angel Pizarro <apizarro@nvidia.com>"
   version: "1.1.0"
   tags:
     - parabricks
@@ -100,6 +101,11 @@ Load only the reference file for the selected tool.
 For routing heuristics when multiple tools could apply, see
 [tool-index.md](references/tool-index.md).
 
+**QC vs variant calling:** `bammetrics` and `collectmultiplemetrics` are metrics
+tools only — they do not fix coverage gaps or call variants. Route variant recall
+goals to callers such as `haplotypecaller`, `germline`, or `deepvariant`, not QC
+tools.
+
 ## Runtime Readiness
 
 For GPU, driver, Docker, container, storage, or installation questions, read